2 ms·

> Deepseekv4 is freely available to download you can host your own and sell it keeping the proceeds and it rivals Claude Opus 4.7. good luck getting a machine

> Deepseekv4 is freely available to download you can host your own and sell it keeping the proceeds and it rivals Claude Opus 4.7. good luck getting a machine that can run its specs though. Even flash is goign to require ponying up 5-10 grand to run the minimal specs for it. The vast majority of people will find their machine falls behind as tech progresses long before they get a return on that investment. That said, it does mean there will be a healthy market for "generic providers" in the AI landscape with these open weight models.

Glohrischi 5mo ago
Companies still pay 5k and more for a basic website. 5-10k is quite affordable. Investment is not basic math. Its also dependencies to US companies, trust etc.
- Lord-Jobo 5mo ago
  Investment and finance in general, at this scale, is far more geopolitics than it is math. It’s self evident. We’ve all seen how the “math” on so much of the AI business sector literally doesn’t check out, and here there are: still ballooning, still making deals, still directly crafting laws through political influence, still taking over damn near every user space. Politics at a high enough level lets you play a different game with different rules. Politics at a low enough level lets you do the same actually, but we usually call that civil unrest, guerrilla warfare, or collective action depending on how many of which group is defying which “rules”. It’s very easy to get used to the guardrails and guidelines around us when they persist and succeed for decades, but they are much more fragile than they appear.
hootz 5mo ago
It's not "there will be", there already is a market of generic providers and you can use millions of tokens of DeepSeek-v4-flash for like 1 dollar. https://models.dev/?search=deepseek-v4-flash https://models.dev/?search=deepseek-v4-flash
- smallmancontrov 5mo ago
  Yeah, but will those hosted models help me write smut, advance my weekend CBRN hobby, advice on how to kill myself, advice on how to kill the person who made me want to kill myself, and how to set up a mega drug manufacturing operation like a real life Walter White?
  - notfromhere 5mo ago
    you don't need AI to deal drugs, plenty of people with little education do it all the time
    - arretevad 5mo ago
      *Formal education
  - gmerc 5mo ago
    You mean Google?
- vablings 5mo ago
  Man, deepseek flash is so damn good, its pricing is insane compared to frontier models and I have found that its performance is about 80-90% of a frontier model for a literal fraction of a cost.
zackify 5mo ago
https://github.com/antirez/ds4 https://github.com/antirez/ds4
Bnjoroge 5mo ago
You only need about a mac w 96GB or 128gb to run deepseek v4flash with ds4(https://github.com/antirez/ds4 https://github.com/antirez/ds4). Works mostly well
- zozbot234 5mo ago
  That's only antirez's 2-bit version though. The real version of DeepSeek V4 Flash will be slow on that machine.
rootnod3 5mo ago
But...so does the tech sector. They will also have to continually upgrade their AI slop data centers to run newer better models, generating a heap of waste along it. And that money has to be made back.
tootie 5mo ago
Can I rent space in Colossus2?
zozbot234 5mo ago
> good luck getting a machine that can run its specs though. That's any machine that can physically host the weights and context. You'd need a highly-specced machine for better performance and throughput, but it's not a requirement as far as literally executing the model and getting output.
gopher_space 5mo ago
I've been taking the opposite approach by playing around with different hardware bottlenecks, and the models I can run on base-level (e.g. 1050 ti, i5-7th) machines are all useful for specific tasks if I deconstruct a process far enough. If you know exactly what you want and when you need the results, calculating a hardware floor becomes deterministic. If you don't need the results right away, or if you're comfortable gluing together results yourself, pretty much any box released in the past decade can be doing work for you. For a quick experiment grab a few images and a one-line prompt e.g. "describe these pictures", and bounce that set off every model/quant you can reach. If you record quality, rate, and cost of each request there might be regions of `"good enough", "good enough", "electricity + sweat"` in the resulting spreadsheet. And this is multimodal. Single-mode classification is dirt cheap.

Glohrischi 5mo ago

Companies still pay 5k and more for a basic website. 5-10k is quite affordable. Investment is not basic math. Its also dependencies to US companies, trust etc.

Lord-Jobo 5mo ago
Investment and finance in general, at this scale, is far more geopolitics than it is math. It’s self evident. We’ve all seen how the “math” on so much of the AI business sector literally doesn’t check out, and here there are: still ballooning, still making deals, still directly crafting laws through political influence, still taking over damn near every user space. Politics at a high enough level lets you play a different game with different rules. Politics at a low enough level lets you do the same actually, but we usually call that civil unrest, guerrilla warfare, or collective action depending on how many of which group is defying which “rules”. It’s very easy to get used to the guardrails and guidelines around us when they persist and succeed for decades, but they are much more fragile than they appear.

hootz 5mo ago

It's not "there will be", there already is a market of generic providers and you can use millions of tokens of DeepSeek-v4-flash for like 1 dollar. https://models.dev/?search=deepseek-v4-flash https://models.dev/?search=deepseek-v4-flash

smallmancontrov 5mo ago
Yeah, but will those hosted models help me write smut, advance my weekend CBRN hobby, advice on how to kill myself, advice on how to kill the person who made me want to kill myself, and how to set up a mega drug manufacturing operation like a real life Walter White?
- notfromhere 5mo ago
  you don't need AI to deal drugs, plenty of people with little education do it all the time
  - arretevad 5mo ago
    *Formal education
- gmerc 5mo ago
  You mean Google?
vablings 5mo ago
Man, deepseek flash is so damn good, its pricing is insane compared to frontier models and I have found that its performance is about 80-90% of a frontier model for a literal fraction of a cost.

zackify 5mo ago

https://github.com/antirez/ds4 https://github.com/antirez/ds4

Bnjoroge 5mo ago

You only need about a mac w 96GB or 128gb to run deepseek v4flash with ds4(https://github.com/antirez/ds4 https://github.com/antirez/ds4). Works mostly well

zozbot234 5mo ago
That's only antirez's 2-bit version though. The real version of DeepSeek V4 Flash will be slow on that machine.

rootnod3 5mo ago

But...so does the tech sector. They will also have to continually upgrade their AI slop data centers to run newer better models, generating a heap of waste along it. And that money has to be made back.

tootie 5mo ago

Can I rent space in Colossus2?

zozbot234 5mo ago

> good luck getting a machine that can run its specs though. That's any machine that can physically host the weights and context. You'd need a highly-specced machine for better performance and throughput, but it's not a requirement as far as literally executing the model and getting output.

gopher_space 5mo ago

I've been taking the opposite approach by playing around with different hardware bottlenecks, and the models I can run on base-level (e.g. 1050 ti, i5-7th) machines are all useful for specific tasks if I deconstruct a process far enough. If you know exactly what you want and when you need the results, calculating a hardware floor becomes deterministic. If you don't need the results right away, or if you're comfortable gluing together results yourself, pretty much any box released in the past decade can be doing work for you. For a quick experiment grab a few images and a one-line prompt e.g. "describe these pictures", and bounce that set off every model/quant you can reach. If you record quality, rate, and cost of each request there might be regions of `"good enough", "good enough", "electricity + sweat"` in the resulting spreadsheet. And this is multimodal. Single-mode classification is dirt cheap.